iii international meeting on plant breedingin this edition the topic addressed were artificial...
TRANSCRIPT
III International Meeting on
Plant Breeding
New Approaches on Plant Breeding: Insights into
Artificial Intelligence
PROCEEDINGS
October – 2019
THE INTERNATIONAL MEETING ON PLANT BREEDING
As the world's population grows, different strategies must be taken by society to
sustainably increase food and supplies production. Plant breeding can be considered one
of the fundamental strategies for the development of more adapted and productive
cultivars. New tools that aid genetic improvement are released every day, aiming to
increase selection accuracy, decrease breeding cycle time and increase genetic gain and,
consequently, productivity. Thus, discussions around these new tools are important to
ensure it correct use and application within research and breeding projects.
The International Meeting on Plant Breeding is one of several events in the
"Corteva Agriscience Plant Science Symposia Series”. In this edition the topic addressed
were artificial intelligence, which has gained the attention of researchers in recent years
for allowing wide application in virtually all stages of plant breeding. It is possible to
apply artificial intelligence to genotypic and phenotypic data, genomic prediction and
mainly image processing. Considering this, the objective of the event was to promote the
discussion of the application of artificial intelligence within the genetic improvement
program.
GVENCK
The Genetics and Plant Breeding Group “Prof. Roland Vencovsky” (GVENCK)
is composed by graduate and undergraduate students in Genetics and Plant Breeding at
“Luiz de Queiroz” College of Agriculture (ESALQ/USP), under coordination of
Professor Dr. José Baldin Pinheiro. Our mission is to integrate academics, professors and
professionals with the goal of improving the training of future breeders and geneticists.
The main activities of the group are:
• Organization of scientific and training events;
• Promotion of discussion on relevant topics in genetic and plant breeding;
• Technical visits to companies and public research institutions;
• Promote the guidance of young talents under training from the “alumni voice”, in which
the alumni with consolidated careers will share professional experiences;
• Promote moments and opportunities for interaction between students, professors and
researchers outside the university;
• Establishment of partnerships with companies and public institutions.
ORGANIZATION
Coordination
José Baldin Pinheiro
Rafael Massahiro Yassue
Ana Letycia Basso Garcia
Secretary
Alline Sekiya
Emanoel Sanches Martins
Jessica Aparecida Ferrarezi
Mariana Niederheitmann
Melina Prado
Pedro Augusto Medeiros Barbosa
Treasurer
Maiara de Oliveira
Patrícia Braga
Scientific
Albania José Patiño Torres
Jéssica Eliana Nogueira de Souza
Renata Flávia de Carvalho
Roberta Luiza Vidal
Talieisse Gomes Fagundes
Communications
Mariana Vianna Bruna de Moura Lopes
Karina Midori Kobayashi Omosako
Marcos Antonio de Godoy Filho
Melissa Cristina de Carvalho Miranda
Yajahaira Nevenka Carbajal Gonzales
Logistics
Guilherme Kenichi Hosaka
Elesandro Bonhofen
Leonardo Fioravante Gotardi
Willian Giordani
ACKNOWLEDGEMENTS
The organizing committee is thankful for all the support provided by Corteva
Agriscience, Bayer, FAPESP (São Paulo Research Foundation), ESALQ (“Luiz de
Queiroz” College of Agriculture), Department of Genetics, The graduate program in
Genetics and Plant Breeding and FEALQ (Foundation of Agrarian Studies Luiz de
Queiroz).
PROGRAM:
October 1st, 2019 - Tuesday
07:30– 8:00 – Registration
08:00 – 08:30 – Opening Session – Tabare Abadie - Corteva Agriscience - USA
08:30 – 09:30 – Lecture 1: André Carlos Ponce de Leon Ferreira de Carvalho – “What is
the relation between Big Data and Data Science? And how does Artificial Intelligence
fits in this relation?”
09:30 - 09:50 – Coffee Break
9:50 – 10:50 – Lecture 2: Alencar Xavier – “Good learners, faster learning”
10:50 – 11:30 – Poster Session
11:30 – 13:30 – Break
13:30 – 14:30 – Lecture 3: Osval A. M. López – “Machine learning and deep learning
methods for genomic prediction”
14:30 – 15:30 – Lecture 4: Katy Martin Rainey – “Phenomic Inference of Soybean
Growth and Development”
15:30 - 16:00 - Coffee Break
16:00 – 17:00 – Lecture 5: Rodrigo G. Trevisan – “Deep learning for image-based high
throughput plant phenotyping”
17:00 – 18:00 – Lecture 6: Vinícius Silva Junqueira – “How Data Science is changing the
on the job experience in industry?”
18:00 – 18:30 – “Roland Vencovsky” Award and Closing Remarks
19:00 – Official Event Dinner
October 2nd, 2019 - Wednesday: Minicourse I
8:00 – 18:00 - Osval A. M. López - “Machine learning methods for genomic prediction”
October 3rd, 2019 - Thursday: Minicourse II
8:00 – 18:00 - Rodrigo G. Trevisan - “A hands-on introduction to deep learning for image-
based high throughput plant phenotyping”
SPEAKERS
Dr. Tabare Abadie has a Ph.D. in Plant Breeding and
Quantitative Genetics from the University of Minnesota.
His areas of interest include Plant Breeding, Higher
Education, Career Development and Mentoring. It is
currently a Corteva Fellow and his main responsibility
is to lead the Plant Sciences Symposia Series. He is also
currently serving as a member of the Board of Trustees
of the Integrated Breeding Platform, and of the
Agronomic Science Foundation (ASF), and as a member
of external scientific advisory panels for USDA and
NSF funded projects at academic organizations.
Dr. André Carlos Ponce de Leon Ferreira de Carvalho
is a Professor at the Institute of Mathematical and
Computer Sciences, University of São Paulo (ICMC-
USP), São Carlos campus. Vice President of the
Brazilian Computer Society (SBC). He holds a
Bachelor degree (1987) and a Master's degree in
Computer Science (1990) from the Federal University
of Pernambuco, and a doctorate in Electronic
Engineering from the University of Kent (1994). His
main research interests are Machine Learning, Data
Mining, and Data Science, with applications in many
areas.
Dr. Alencar Xavier has his Ph.D. degree in soybean
breeding and genetics statistics at Purdue University.
His works on statistical genetics are focused on
genomic-assisted breeding with emphasis on theoretical
and computational aspects of data-driven plant
breeding, such as modeling, prediction, and selection
using various sources of information. His researches
regard the development and implementation of new
quantitative methods using mixed models, Bayesian
methods and machine learning, along with high-
performance computing. Currently, hi is a quantitative
scientist at Corteva Agriscience.
Osval A Montesinos-López has PhD. in statistics and
biometry and is a professor at the Faculty of Telematic,
Universidad de Colima in Mexico. Dr. Osval researches
are currently in Biostatistics and Bioinformatics.
Dr. Katy Martin Rainey is an Associate Professor of Plant
Breeding and Genetics in the Purdue University Department
of Agronomy. She studies the genetics improvement of
soybeans for increased yield and better quality using
multidisciplinary approaches.
Rodrigo Trevisan is a graduate research assistant in
Crop Science at the University of Illinois at Urbana-
Champaign - USA. He is currently responsible for
research and development of technology solutions for
agribusiness at Smart Agri. Has experience in
agricultural planning, precision agriculture, farm level
experimentation, remote sensing, geographic
information systems, data analysis, and artificial
intelligence.
Dr. Vinícius Junqueira holds a degree in veterinary
medicine from the University of Brasilia (2010), a
master's degree in Genetics and Breeding at the Federal
University of Viçosa (2014) and a doctorate in Genetics
and Breeding at the Federal University of Viçosa (2018).
He was a visiting researcher at the University of Georgia
(2016). He is currently Data Science Supervisor - Bayer
Corporation.
SUMMARY
COMPARISON OF TWO APPROACHES FOR CONTIG ORDERING IN GENOME
ASSEMBLIES .................................................................................................................. 1
FACTOR ANALYSIS APPLIED TO MAIZE ADAPTABILITY STUDIES AND
ENVIRONMENTAL STRATIFICATION ...................................................................... 2
GENOMIC X ENVIROTYPING KERNELS DRIVE TO A BETTER PREDICTION
AND UNDERSTANDING OF MAIZE YIELD PLASTICITY ...................................... 3
IMPACTS OF PAST AND FUTURE CLIMATE CHANGE ON GENETIC
DIVERSITY OF TWO Eremanthus SPECIES ................................................................ 4
KINSHIP COEFFICIENTS IN FOREST SPECIES ........................................................ 5
MACHINE LEARNING-ASSISTED IMAGE ANALYSIS TO QUANTIFY FOLIAR
DISEASE SEVERITY FROM TOMATO BACTERIAL SPOT ..................................... 6
MULTI-LAYER PERCEPTRON (MLP) NETWORKS AND DIALLEL ANALYSIS
APPLIED TO MAIZE POPULATION SELECTION ..................................................... 7
SOYBEAN LINEAGES TOLERANT TO ASIAN SOYBEAN RUST .......................... 8
STUDY OF AUSTROPUCCINIA PSIDII DURING INFECTION IN EUCALYPTUS
GRANDIS REVEALS MEIOTIC-SPECIFIC FEATURES OF THE PATHOGEN ........ 9
WITHIN-FAMILY GENOMIC SELECTION IN SUGARCANE FOR BROWN RUST
RESISTANCE VIA MACHINE LEARNING ............................................................... 10
1
COMPARISON OF TWO APPROACHES FOR CONTIG ORDERING IN
GENOME ASSEMBLIES
Marcos Antonio de Godoy Filho1; Lucas Borges dos Santos2; João Paulo Gomes Viana3;
Wei Wei3; Steven J. Clough3,4; and José Baldin Pinheiro1
1Escola Superior de Agricultura Luiz de Queiroz, Piracicaba, SP. 2Universidade Estadual
de Campinas. 3University of Illinois at Urbana-Champaign, 4US Dept of Agriculture,
Urbana, IL. E-mail: [email protected].
Soybean is the main commodity in Brazil, in 2018, soybean and its derivatives accounted
for $40.9 billion in exports. However, there are some obstacles to obtaining higher yield,
and among the difficulties, the attack of insects stands out, especially by the complex of
stink bugs. An alternative to the use of insecticides is the development of resistant
cultivars. The assembly and comparison of contrasting genomes can be a good approach
to identify polymorphisms between them, and allows integration with differentially
expressed gene data and QTLs for a better understanding of the genetic architecture of
the trait. However, genome assembly and ordering are challenges considering the high
complexity of soybean genome. For a better understanding of the genetic architecture
involved in the resistance of soybean to the stink bug complex, genomes of soybean
cultivar IAC-100 (tolerant) and cultivar CD-215 (susceptible) were sequenced. For this,
we used the Chromium Genome technology, and sequencing was performed on the
Illumina NovaSeq 6000 plataform. The draft genome assembly was performed using the
software Supernova 2.1.1, which for the IAC-100 generated an assembly with 1.051MB
with a scaffold N50 of 4.02 Mb. For the cultivar CD-215, an assembly of 1.055MB was
generated, with an N50 scaffold of 4.28 MB. Two approaches for ordering and breaking
chimeric contigs were tested, using Ragoo based on reference genome of Willians 82 and
Cromonomer based on consensus genetic map of soybean. Contig ordering and breaking
proved to be highly efficient in increasing the quality of genome assembly, with Ragoo
improving the N50 scafolds up to 53.5MB in IAC100 and 53.8MB for CD215, plus break
44 chimeric contigs in assembly of IAC100 and 49 in CD215. This result for N50 scaffold
is better than all soybean genomes available in NCBI, in addition CG content is close to
the other assemblies available (~31%) and 75% of assembly of both genomes are only 15
scaffolds. The Cromonomer boosted N50 scafolds to around 35-37MB but found fewer
chimeric contigs, however, it should be noted that just over 2700 SNPs lined up against
assemblies and was used for ordering using the genetic map. The comparative genomics
through genome assembly in soybeans opens the door to several other analyzes and
studies in order to understand the genetic architecture of bed bug complex tolerance. The
combination of supernova and Raggo softwares can be a powerful tool for assembling
genomes that already have a reference, with Chromonomer software being a great
alternative for species without assembled genomes but with good genetic maps.
Keywords: Assembly; Soybean; Genome; Ordering; Stinck Bugs.
2
FACTOR ANALYSIS APPLIED TO MAIZE ADAPTABILITY STUDIES AND
ENVIRONMENTAL STRATIFICATION
Maria Angélica Marçola1; Yasmin Moura Araújo2; Deoclécio Domingos Garbuglio3;
Gabriela Inocente4; Pedro Mário de Araújo5
1Instituto Agronômico do Paraná (IAPAR); 2 Agronômico do Paraná (IAPAR); 3Instituto
Agronômico do Paraná (IAPAR); 4Universidade Estadual de Londrina (UEL); 5Instituto
Agronômico do Paraná (IAPAR). E-mail: [email protected].
For adaptability and stability studies, maize cultivars are evaluated in different regions
with edaphoclimatic diferences in order to explore the effects of genotype x environment
interaction (GxE). In this context, among univariate methods, linear regression has been
widely used, but is not informative if linearity fails and tends to simplify response models.
Among the multivariate techniques, factor analysis show promising results, being used in
plant breeding relatively recently, both in environmental stratification and genotype
adaptability studies. Thus, the objectives of this work were to evaluate the adaptability
and stability of genetically modified maize cultivars and to perform environmental
stratification in the IAPAR maize evaluation network, using the factor analysis technique.
The trials were carried out 2018/2019 crop, in the location: Londrina, Santa Tereza do
Oeste, Santa Helena, Palotina, Cambará, Foresta, Campo Mourão, Pato Branco,
Guarapuava and Ponta Grossa, evaluating 13 maize cultivars. The experimental design
was randomized complete blocks with three replications per evironment, and the plots
consisted of two rows of 5 meters length, 80 cm between rows and 20 cm between plants.
Through the yeld means, were identified the cultivars FS 481 PW and 2B500 PW, which
showed wide broad adaptability to different environments clutered into 4 final factors.
Cultivars R9789-VIP3 and R9080-PRO2 presented low adaptability in factor sets F1xF2,
F1xF3 and F1xF4, which clusted 53.8%, 38.4% and 38.4% of the environments, within
each combination of factors, respectively. The other cultivars presented specific
adaptability to certain factor combinations. The factor analysis technique is efficient for
the identification of genotypes with different responses to GxE interaction, as well as for
grouping of environments.
Keywords: Zea mays L.; GxE interaction; Multivariate analysis; Heatmaps.
3
GENOMIC X ENVIROTYPING KERNELS DRIVE TO A BETTER
PREDICTION AND UNDERSTANDING OF MAIZE YIELD PLASTICITY
Germano Martins F. Costa-Neto1; Giovanni Galli1; Roberto Fritsche-Neto1
1Luiz de Queiroz College of Agriculture, University of São Paulo (ESALQ/USP), E-mail:
Whole-genome prediction (WGP) is guided from several genotype-to-phenotype (G-to-
P) modeling strategies integrated into a time-saving, cost-effective breeding pipeline.
However, over multi-environment analysis, the genotype by environment interaction
(GE) and their unpredictable impact on phenotypic plasticity may limit the range of
applications of WGP-based breeding. One of the strategies used to solve this gap relies
on the integrative envirotyping-based data with WGP modeling, which are focused on
emulating the latent genotypic-specific reaction norms across the mean environmental
gradient. Here we integrate explicit envirotyping data with WGP by emulating the
implicit envirotype-to-phenotype (E-to-P) dynamics driving GE effects and yield
plasticity over a target population of environments (TPE) in maize. From available
phenotypic data of an optimized training population set (N = 50% of genotypes), we
obtained the empirical environmental relationship kernel (*KE), which is trait-specific for
grain yield. Then, partial least squares regression with envirotyping data (e.g., daily
weather, nitrogen level) were used to translate the trait-specific E-to-P dynamics present
in *KE into a realized environmental kernel KE. One advantage is that we can fit an
approximation of the infinitesimal environment effect and compute its impact on the
overall environmental quality driving KE for a specific TPE. Finally, genomic best-
unbiased predictions (GBLUP) were carried out, incorporating additive and dominant
effects with the KE-based reaction norm. As a proof-of-concept, we computed the
predictive ability (r), selection coincidence (SC) and root mean squared error prediction
(RMSE) front to the benchmark additive-dominant single-variance GE deviation
GBLUP, over 5-folds of leave-one-environment and 3 prediction scenarios: (i) new GE
(nGE); (ii) new genotypes (nG) and (iii) new environments (nE). We observed lower r
gains in the scenarios nGE (+5%) and nE (+3%), but an increased by 10% for nG,
indicating that the modeling focused on target trait and TPE may improve prediction for
new genotypes within the multi-environment testing network. However, SC% values
were increased for all scenarios, from 10% (nGE) to 15% (nG), and RMSE values reduced
by up 391%. In addition, we identified the orthogonal impact of the environmental
sources onto GE variation, such as solar nitrogen level (20%), radiation across crop
development (50%), wind speed (35%) and air temperature at initial growing stages
(30%). Our results indicate that genomic × envirotyping based kernels better emulates the
genotypic ranks for target environments. Therefore, the quantitative and qualitative
differences between the genotypic-specific plastic responses may be better modeled and
exploited for the whole TPE. Finally, the genomic × envirotyping modeling expands the
WGP spectrum of applications, such as for yield adaptability screening or for design
future breeding strategies.
Keywords: genomic selection; reaction norm; multi-environment; hybrid testing.
4
IMPACTS OF PAST AND FUTURE CLIMATE CHANGE ON GENETIC
DIVERSITY OF TWO Eremanthus SPECIES
Lucas Fernandes Rocha1,2; Isaias Emilio Paulino do Carmo1,3; Joema Souza Rodrigues
Póvoa1; Dulcinéia de Carvalho1; Evandro Vagner Tamburissi2,4
1Departamento de Ciências Florestais, Universidade Federal de Lavras, Brazil. 2Departamento de Ciência Florestal, Universidade Estadual Paulista “Júlio de Mesquita
Filho”, Brazil. 3Institute of Biosystems Engineering, Poznan University of Life Sciences,
Poland. 4Departamento de Engenharia Florestal, Universidade Estadual do Centro-Oeste,
Brazil. E-mail: [email protected].
Phylogeographic patterns of endemic species is a critical key to understand its adaptation
to future climate change. Here, based on chloroplast DNA (cpDNA), we analyzed the
genetic diversity of two endemic and endangered tree species from the Brazilian savanna
and Atlantic forest (Eremanthus erythropappus and Eremanthus incanus). Additionally,
we applied the climate-based ecological niche modeling (ENM) to evaluate the impact of
the Quaternary climate (Last Glacial Maximum ~ 21 kyr BP and Mid-Holocene ~ 6 kyr
BP) on the current haplotype distribution. Moreover, we modeled the potential effect of
future climate change on the species distribution in 2070 for the most optimistic and
pessimistic scenarios. One primer/enzyme combination (SFM/HinfI) revealed
polymorphism with very low haplotype diversity, showing only three different
haplotypes. The haplotype 1 has very low frequency and it was classified as the oldest,
diverging from six mutations from the haplotypes 2 and 3. The E. erythropappus
populations are structured and differ genetically according to the areas of occurrence. In
general, the populations located in the north region are genetically different from those
located in the center-south. No genetic structuring was observed for E. incanus. The ENM
revealed a large distribution during the past and a severe decrease in geographic
distribution of E. erythropappus and E. incanus from the LGM until present and predicts
a drastic decline in suitable areas in the future. This reduction may homogenize the
genetic diversity and compromise a relevant role of these species on infiltration of
groundwater.
Keywords: ecological niche modeling; genetic diversity; climate change; chloroplast
DNA.
5
KINSHIP COEFFICIENTS IN FOREST SPECIES
Antônio Higo Moreira de Sousa1; Lucas Fernandes Rocha1; Evandro Vagner
Tambarussi1,2
1Departamento de Ciência Florestal, Universidade Estadual Paulista “Júlio de Mesquita
Filho”, Brazil.2Departamento de Engenharia Florestal, Universidade Estadual do Centro-
Oeste, Brazil. E-mail: [email protected].
Understanding kinship between individuals in natural populations offers useful
information that can be assessed based on estimators of molecular data. However, each
estimation method is based on assumptions and is often restricted for forest species. Thus,
modeling the data is an important step that is almost always neglected. This study aims
to evaluate the efficacy of different kinship estimators for Acrocomia aculeata (Jacq.)
Lodd, ex Mart., Hymenaea stigonocarpa Mart. Ex Hayne, and Dipteryx alata Vogel., and
four different simulated populations. From the estimated mean coancestry (�̅�), the error
estimates were calculated assuming that analyzed individuals were perfect half-siblings
(�̅�=0,125). Estimates of kinship ranged according to the species and method used,
generating different error values for each estimator. A correlation was observed only
between estimators that used the same estimation method or similar assumptions.
Simulated populations showed more accurate estimates and lower error values compared
to actual data. The error values of the estimators demonstrate that the application of
estimators to infer a certain degree of kinship can generate biased results and lead to
inefficient decision making. Thus, the use of complementary information associated with
kinship is necessary, such as analysis of the reproduction system and genetic structure of
the population, enabling more precise inferences of the kinship between evaluated
individuals.
Keywords: kinship estimators; forest species; genomic selection; simulations; mixed
reproductive system.
6
MACHINE LEARNING-ASSISTED IMAGE ANALYSIS TO QUANTIFY
FOLIAR DISEASE SEVERITY FROM TOMATO BACTERIAL SPOT
Mariana Niederheitmann1; Jéssica E. Nogueira de Souza1; Daniel P. Longatto1; Maria
Carolina Quecine Verdi1; Fernando A. Piotto2; David M. Francis3
1‘Luiz de Queiroz’ College of Agriculture, Genetics Department, University of São Paulo,
Brazil. 2‘Luiz de Queiroz’ College of Agriculture, Crop Science Department, University
of São Paulo, Brazil. 3Ohio Agricultural Research and Development Center, Horticulture
and Crop Science Department, The Ohio State University, Wooster, USA. E-mail:
Foliar disease severity in plants is often assessed using rating or diagrammatic scales.
Scaled evaluations may not fully capture genetic signals and the nonparametric features
may limit their effective use in genome-wide association studies and fine mapping
approaches, which assume a continuous distribution. Image analysis, utilizing
segmentation based on machine learning algorithms, provides a straightforward tool for
obtaining quantitative information of disease severity. Foliar symptoms of bacterial spot
(BS), a major disease in many tomato-growing regions worldwide, include lesions at the
margin of leaves and water-soaked, brown, angular spots. These patterned symptoms
contrast with leaf and background color, which can be recognized by pixel classification
algorithms. In this study, various methods used for image segmentation were tested to
assess BS symptoms in 466 leaf images from 20 tomato genotypes displaying variable
disease severities and previously rated with a commonly used scale. The leaves were
classified and measured for disease severity using the Trainable Weka Segmentation and
Color Pixel Counter plugins, respectively, both available in the free and open
environment software ImageJ/Fiji. Methods were assessed based on computational time,
correlations with rating data, and ability to partition variance into genetics (heritability).
Results revealed high heritability (up to 35,0%) and significant correlation of some of the
tested methods with disease severity observed in genotypes with known resistance. The
most predictive methods were those in which the training image was DSCN9921, which
was likely more representative for classifying the other images. The number of selections
for background, healthy tissue and diseased tissue were not significantly associated with
improved heritability or correlation values, though for some analysis 16-18 appear
optimal. Adding the number of training algorithms beyond 3 did not improve prediction
and indicated to affect running time. These results point out the potential of such
automated tools for determination of disease severity in genetics and crop improvement
studies.
Keywords: pixel classification; Xanthomonas perforans; plant breeding; Solanum
lycopersicum.
Acknowledgements: CAPES, CNPq, Ohio Agricultural Research and Development
Center-The Ohio State University (OARDC-OSU).
7
MULTI-LAYER PERCEPTRON (MLP) NETWORKS AND DIALLEL
ANALYSIS APPLIED TO MAIZE POPULATION SELECTION
Gabriela Inocente1; Deoclécio Domingos Garbuglio2; Pedro Mario de Araújo3; Maria
Angélica Marçola4
1Universidade Estadual de Londrina (UEL); 2Agronomic Institute of Paraná (IAPAR);
3Agronomic Institute of Paraná (IAPAR); 4Agronomic Institute of Paraná (IAPAR) E-
mail: [email protected].
In the maize breeding programs the search for commercial hybrids may initially be
concentrated on the identification of base populations for inbred lines extraction. These
populations are direct to crosses in which the best interpopulation hybrids can be used as
an indication to justify the extraction of lines with good allelic complementarity. Thus,
the present work aimed to evaluate a complete diallel, according to the methodology
diallel analysis proposed by Eberhart and Gardner (1966) with adaptation of Morais et al.
(1991), between 12 maize populations and predict, through a multi-layer perceptron
(MLP) neural network model, selection patterns based on mean components. Twelve
intercrossed maize populations were evaluated at the IAPAR (Agronomic Institute of
Paraná) experimental stations in three locations: Londrina, Guarapuava and Ponta Grossa,
Paraná, during the 2005/2006 crop. The experimental design was randomized complete
blocks with two replications per location. The experimental plot was composed of a row
with five meters long and five plants per meter after thinning. To identify the population
base, only the variable grain yield (GY) was evaluated. It was found that the implemented
MLP architecture (one input layer, three hidden layers and one output layer) and other
associated parameters, allowed the development of a generalist model of genotype
classification (Classes A, B, C and D), based on diallel model (General Combining
Ability and Specific Heterosis). Thus the MLP neural network model was efficient in
predicting selection classifications from mean components obtained from a complete
diallel, regardless of the evaluation environment.
Keywords: Zea mays L., Artificial Neural Networks, Intermediate Hybrids, Diallel
Crosses.
8
SOYBEAN LINEAGES TOLERANT TO ASIAN SOYBEAN RUST
Moreno, B.M1; Martins, G.M.1; Stella, A.A.1; Nekatschalow; M.C.1; Vello, N.A.1;
Pinheiro, J.B1.
1 Genetics Department, University de Sao Paulo - “Luiz de Queiroz” College of
Agriculture USP/ESALQ. E-mail: [email protected].
Asian soybean rust is one of the most important diseases that affect soybean production
around the world. The season 18/19, FRAC identified a sensibility reduction in sporus
collected and after exposed to compounds such as triazole and triazolinthione. As soybean
occupies around 36 million hectares, and the pathogen Phakopsora pachyrhizi shows high
rates of mutation; using tolerant cultivars can be promising disease management. The
main objective was to evaluate and identify in 256 experimental soybean lineages in F5:8,
tolerance to soybean rust using two treatments (one of them controlling the disease
complex of the end of the cycle and the soybean rust, and the other one, controlling the
end-cycle diseases and without the control) using the rust effect approach. The experiment
was conducted in Piracicaba/SP, in the summer season of 18/19, using Federer blocks as
experimental design. Each treatment had two repetitions, to evaluate 256 experimental
lineages and 3 controls (IAC100, BRS133 and CB 07-958-B, commercial cultivars). Each
block had 1 m2. The plants were evaluated in plant height, the number of days to maturity,
yield, and weight of 100 seeds. In addition, the rust effect was estimated based on the
difference between the treatment with the rust control and the treatment without the rust
control.The mean yield of lineages with the rust control was 3.827,47 kg.ha-1, and without
the control was 3.255,61 kg.ha-1. The weight of 100 seeds was 17,26 g and 13,82 g
respectively. The plant height, 97 cm, and 93 cm. And, the plant cycle, 136 days and 127
days. The mean rust effect calculated for yield was 0,571 kg.ha-1, for the weight of 100
seeds was 3,44 g, height 4 cm and cycle 8,7 days. In conclusion, the plant evaluation has
shown that 22 lineages present desirable yield, even when exposed to the disease, and
have good agronomic traits. Then, these lineages were chosen as the best lineages to
continue the breeding program. This lineages are USP 17-42.101, USP 17-42.121, USP
17-42.421, USP 17-42.111, USP 17-42.073, USP 17-42.043 and USP 17-42.223. And the
most tolerant experimental lineages to soybean rust are USP 17-42.143, USP 17-42.101,
USP 17-42.121, USP 17-42.284, USP 17-42.133, USP 17-42.102, USP 17-42.231, USP
17-42.592, USP 17-42.034, USP 17-42.222, USP 17-42.213, USP 17-42.074, USP 17-
42.413, USP 17-42.402 e USP 17-42.103.
Keywords: Glycine max; Phakopsora pachyrhizi, Genetic tolerance; Rust effect.
9
STUDY OF AUSTROPUCCINIA PSIDII DURING INFECTION IN
EUCALYPTUS GRANDIS REVEALS MEIOTIC-SPECIFIC FEATURES OF
THE PATHOGEN
J. A. Ferrarezi1; C. A. A. Hayashibara1; T. B. de Lima1; M. da S. Lopes1; M. C. Quecine1
1Department of Genetics, College of Agriculture “Luiz de Queiroz”, University of Sao
Paulo, Piracicaba-SP, 11 Padua Dias Avenue, 13418-900, Brazil. E-mail:
The basidiomycete fungus Austropuccinia psidii is the causal agent of myrtle rust, which
may infect approximately 270 Myrtaceae species. Among those, Eucalyptus spp. stand
out as one of the most economical important genus attacked by this pathogen worldwide.
A. psidii is a biotrophic pathogen and little is known about its biological and genetic
aspects. We investigated the presence of meiotic genes in the pathogen’s genome and
transcriptome, and monitored meiosis-related structures by inoculating urediniospores of
A. psidii in E. grandis and analysing a 52-day-long time course. Comparative analysis of
136 genes present in Ustilago maydis, a model basideomycete to sexual cycle, with A.
psidii and other 4 rust species resulted in 116 genes common for all compared genomes.
We found 123 meiotic genes in A. psidii draft genome, including cell cycle control,
condensation and chromosome structure maintaining, DNA replication, chromatids
maintaining protein, DNA and incorrect pairing repair, double strand break formation,
strand invasion, synapse and recombination. Furthermore, genes related to sporulation
and teliospore germination were found in A. psidii’s genome. Spore samples were
collected at 0, 14 and 28 days after inoculation (DAI) for microscopic and gene expression
analysis. We investigated the expression of 3 mitotic genes and 3 meiotic genes. We found
teliospores during interaction with E. grandis and it is an indicator of the starting-point
of the pathogen’s meiosis. Relative expression ratios of the mitotic-related genes DNA
helicase and G2-specific cyclin cyb were significantly down-regulated, while two meiotic
genes – CDK assembly factor and checkpoint DNA exonuclease rad1 - were significantly
up-regulated in samples of spores collected 28 DAI in E. grandis compared to 14 DAI,
corroborating with the difference observed in teliospore abundance during time course
monitoring. This clear evidence supports the hypothesis that A. psidii is able to undergo
meiosis in teliospores produced during interaction with E. grandis. We conclude that A.
psidii teliospores are produced in very low rates at early stages of rust in E. grandis and
expressively increases around 50 DAI, basidiospores were observed after teliospore
germination in vitro, as never described before in literature.
Key words: gene expression; myrtle rust; meiosis; mitosis; teliospore.
Acknowledgements: This work was supported by CNPq, CAPES and FAPESP.
10
WITHIN-FAMILY GENOMIC SELECTION IN SUGARCANE FOR BROWN
RUST RESISTANCE VIA MACHINE LEARNING
Alexandre Aono1; Ricardo Pimenta1; Melina Mancini1; Cláudio Cardoso-Silva1;
Fernanda dos Santos2; Luciana Pinto2; Reginaldo Kuroshu3; Anete Pereira de Souza1
1Centro de Biologia Molecular e Engenharia Genética, Universidade Estadual de
Campinas, Campinas, Brazil. 2Centro de Cana, Instituto Agronômico de Campinas,
Ribeirão Preto, Brazil. 3Instituto de Ciência e Tecnologia, Universidade Federal de São
Paulo, São José dos Campos, Brazil. E-mail: [email protected].
Sugarcane production is a major national source of income due to its bioenergetic
potential. One of the most common pathogens affecting this crop is brown rust (Puccinia
melanocephala), which has spread worldwide in growing areas generating considerable
economic impact. Scientific advances concerning the genetic resistance of sugarcane to
this disease have been limited because of its large, complex and polyploid genome.
Besides, these studies face the lack of appropriate methodologies, softwares and genomic
references. A full-sib progeny with 144 individuals from Agronomic Institute of
Campinas sugarcane breeding program was used to develop a method for the
identification of varieties resistant to brown rust. The progeny was genotyped using
Genotyping-by-Sequencing. High quality reads were mapped to 7 genomic references.
Five tools were used for SNP calling; the final SNP set was obtained from the intersection
criteria of at least two of them. Due to sugarcane’s aneuploidies, the reference allelic
proportions of SNPs were used for analyses. Phenotypic data from 5 years of rust
symptom scoring were evaluated using ANOVA and Tukey’s test. Individuals were
clustered in groups based on rust severity using several statistics methods; a T-test was
performed to check for differences among groups. Using a leave-one-out validation
strategy, SNPs were used to predict severity groups via machine learning (ML) with
various algorithms (K-Nearest Neighbors, Support Vector Machine, Decision Tree,
Random Forests, Multilayer Perceptron, AdaBoost and Gaussian Naive Bayes) and
feature selection strategies. In the final set, 14,540 putative SNPs and 2 phenotypic groups
with considerable differences on rust severity were identified. The first models built using
ML algorithms achieved a mean accuracy of 52%. However, using feature selection
strategies we could reduce our dataset to 136 SNPs achieving a mean accuracy of 72%,
and a maximum of 95% in the most promising algorithm. These results show the great
potential of ML strategies to evaluate phenotypes in highly complex polyploids. Through
the identification of non-linear relationships among genomic regions, these
methodologies can auxiliate sugarcane genomic studies and be used for genomic selection
in breeding programs to predict potential phenotypes.
Keywords: GBS, Polyploidy, Aneuploidy, Classification.
Acknowledgements: We acknowledge FAPESP and Capes for funding.
ORGANIZATION
SUPPORT